Add LLM token

backend/app/api/routes/compliance.py

@@ -85,9 +85,10 @@ async def analyze_stream(
     Events: stage | source | finding | done | error
     """
     from app.application.compliance.pipeline import (
+        detect_cross_clause_conflicts,
         extract_text_from_doc_id,
         extract_text_from_file,
-        run_clauses_parallel,
+        run_clauses_streaming,
         split_into_clauses,
         synthesize_conclusion,
     )
@@ -135,23 +136,27 @@ async def analyze_stream(
             await asyncio.sleep(0)
             clauses: list[str] = await asyncio.to_thread(split_into_clauses, para_text, client)
 
-            # ── Stage 3: retrieve + gap check (parallel across all clauses) ────────────
+            # ── Stage 3: progressive per-clause retrieve + gap check ──────
             findings: list[dict] = []
+            total_clauses = len(clauses)
 
             yield _sse({
                 "type": "stage",
                 "stage": "analyzing",
-                "label": f"Analyzing {len(clauses)} clauses in parallel…",
+                "label": f"Analyzing {total_clauses} clauses…",
             })
+            # Emit initial progress so the frontend can show the total count
+            yield _sse({"type": "progress", "done": 0, "total": total_clauses})
             await asyncio.sleep(0)
 
-            clause_results = await run_clauses_parallel(
+            done_count = 0
+            # Stream results as each clause completes (not after all finish)
+            async for res in run_clauses_streaming(
                 clauses, retrieval_service, client,
                 top_k=5,
                 domains=domains or None,
-            )
-
-            for res in clause_results:
+            ):
+                done_count += 1
                 i = res["index"]
                 chunks = res["chunks"]
                 finding = res["finding"]
@@ -165,14 +170,25 @@ async def analyze_stream(
                         "score": round(float(getattr(chunk, "score", 0)), 3),
                         "status": "retrieved",
                         "full_content": (getattr(chunk, "text", "") or "")[:300],
+                        "clause_index": i,
                     })
 
                 if finding:
                     findings.append(finding)
                     yield _sse({"type": "finding", **finding})
 
+                # Real progress update after each clause completes
+                yield _sse({"type": "progress", "done": done_count, "total": total_clauses})
                 await asyncio.sleep(0)
 
+            # ── Stage 3b: cross-clause conflict detection ─────────────────
+            if findings:
+                conflicts = await asyncio.to_thread(
+                    detect_cross_clause_conflicts, findings, client
+                )
+                if conflicts:
+                    yield _sse({"type": "conflicts", "items": conflicts})
+
             # ── Stage 4: synthesize conclusion ────────────────────────────
             yield _sse({"type": "stage", "stage": "concluding", "label": "Generating conclusion…"})
             await asyncio.sleep(0)